Health care policymakers and practitioners are increasingly pointing to overprescribing as a critical issue in the U.S. health care delivery system. Much of the attention has centered on opioids, as years of high prescribing along with other factors have culminated in a crisis. But the picture is equally concerning for other drug classes. For example, benzodiazepines drive a significant fraction of overdose hospitalizations, and overprescribing of antibiotics is raising the threat of bacterial resistance.1,2
The scale of the problem requires effective, evidence-based solutions to reduce overprescribing. Historically, the standard approach has been to facilitate intensive education, training, and a review of physicians’ prescribing practices. While these methods can move the needle, they are often expensive and hard to scale up.3 Other classic approaches include the use of audit and feedback techniques and computer-generated alerts to provide information to practitioners about their recent performance. Unfortunately, although systematic reviews have shown that these methods can have positive effects , the impacts tend to be small in magnitude.4,5
This body of evidence has left a gap between the need to stem overprescribing quickly and cheaply and the evidence-based tools available to accomplish this goal. Fortunately, evidence from new research suggests that simple interventions can be scalable, low cost, and effective for reducing overprescribing. In the present article, we first review recent light-touch “nudge” interventions on prescribing. We then discuss two recent studies, one of which we conducted, that deployed strong and surprising messages through letters to improve prescribing quality.
Nudging Physicians Toward Better Prescribing
The broad influence of behavioral economics has inspired many health care researchers to test behaviorally informed, nonfinancial approaches to change physician behavior. This work is motivated by evidence that people use a common set of shortcuts in order to make decisions. Interventions that exploit these shortcuts to change behavior without restricting an individual’s choice are often called “nudges.”6,7
Examples of such nudges have included displaying commitment posters in physicians’ offices, sending outlier letters to the highest-prescribing 20% of physician practices, and changing default prescription settings in electronic health records.8–11 Such strategies commonly take advantage of social and professional norms by highlighting to physicians that their prescribing practices are unusual or conflict with clinical guidelines. Another powerful approach is to shift a process (e.g., the default supply on a prescription order) to make the most clinically appropriate approach the easiest for physicians to follow.
One unresolved issue in the field of behavioral economics is the question of which interventions are most effective. This issue was carefully examined in an influential study that compared three behaviorally informed approaches to reducing inappropriate antibiotic prescribing: peer comparison emails, a prompt in the medical record that suggested alternatives to antibiotics, and a prompt that required physicians to enter a reason for their prescription.12 While the suggested-alternatives approach did not have a statistically significant impact on prescribing, the other two interventions led to statistically and clinically meaningful (>20%) reductions in inappropriate prescribing. Moreover, the peer comparison emails continued to reduce inappropriate prescribing a year after the intervention ended.13
A single study cannot definitively establish that one behavioral approach is consistently better than others. The research described above demonstrates that even specific nudges such as changing the default settings for prescription drug orders can perform well in some scenarios (e.g., substituting generic medications for brand medications) and can fail to perform in others (e.g., presenting alternative orders for antibiotics).10,11,13,14 There are many variations for behavioral interventions, all of which may have differing effects depending on local context. The science of how to account for these factors is still developing, and, because many nudges can be low cost or even free to implement, they are still likely to be worthwhile avenues for many organizations.
Recent Effective Interventions Against Overprescribing
Two recent studies took a different tack to improve the value and safety of prescribing that moved beyond a simple nudge: sending physicians letters with surprising or even shocking information. These letters aimed to get physicians’ attention and motivate them to change their prescription practices.
In one of these studies, which we led and participated in, we worked with the Centers for Medicare and Medicaid Services (CMS) and the Office of Evaluation Sciences to test letters with the goal of reducing prescribing of the antipsychotic drug quetiapine (branded as Seroquel® and Seroquel XR®).15 Many quetiapine prescriptions are unnecessary or dangerous, particularly for people with dementia, with guidelines strongly discouraging prescriptions in such cases.16–18 We identified about 5,000 primary care physicians who were high prescribers of quetiapine. Half of the doctors were randomly selected to receive “treatment” letters emphasizing that they had prescribed an unusually high amount of quetiapine and that this behavior had placed them under review by Medicare. The other half of physicians received bland “placebo” letters that discussed a policy that had nothing to do with antipsychotics.
The quetiapine letters caused substantial and long-lasting reductions in prescribing (16% over 2 years). Doctors cut back on prescriptions across the board, including for patients who were both appropriate and inappropriate candidates for the drug, raising the question of whether the letters curtailed beneficial prescribing. However, patients who were more inappropriate candidates had bigger reductions than those who appeared likely to benefit from the drug. Despite this reduction in prescribing, patients did not have observably worse health outcomes, although our ability to detect adverse events was somewhat limited.
The other recent study, led by Jason Doctor at the University of Southern California, tracked physicians and other practitioners who had prescribed opioids to patients who subsequently overdosed and died.19 The San Diego County Medical Examiner Office then sent a letter to the prescribers of a random half of these deceased patients. The letter told the prescribers that their patient had died of an overdose after having received opioids from them. The prescribers of the remaining patients served as controls and were not sent any letter as part of the study.
The overdose letters had a meaningful effect, resulting in a 10% reduction in opioid prescribing over 3 months. Prescribers became less likely to initiate opioids to new patients, and they also cut back on high-dose, potentially dangerous opioid prescribing. Because the study did not directly evaluate patient outcomes, we cannot say for certain that the letters made opioid prescribing safer for the patients of these doctors. For example, if the letters led to rapid or careless reductions in prescribing for long-term opioid users, they could have caused physical and emotional harm if there was no plan for how to replace the effect of opioid therapy. Still, given the link between opioid initiation, high-dose prescribing, and adverse outcomes, the results suggest that the letters may have improved the quality of opioid prescribing.20,21
Features Contributing to the Effectiveness of the Interventions
How did the quetiapine and opioid letters create such meaningful impacts — and what lessons do they provide for practitioners seeking to improve the value of care? While more research is needed to definitively establish the mechanisms of action, we believe that the content of the letters made a big difference. Drawing on insights from behavioral science, we believe that three features of the letters may have increased the likelihood that the intervention had a meaningful effect, as described below.
The quetiapine letters compared physicians with their geographic peers, a classic nudge approach that borrows from the behavioral economics and science literature. Showing someone that their behavior is unusual relative to their peers can provide a powerful impetus for change; for example, peer comparisons can encourage people to pay their taxes and to conserve energy.22,23
However, there is mixed evidence that a simple peer comparison message is enough to drive large changes in prescribing practices. While the previously mentioned peer comparison emails durably reduced inappropriate antibiotic prescribing by >20%,13 other audit and feedback studies involving peer comparisons have yielded only small improvements in clinical quality. For example, two recent studies demonstrated that peer comparison letters led to no changes in the prescribing of controlled substances and opioids,24,25 and another study (mentioned earlier) demonstrated that peer comparison letters led to a meaningful but small 3% decline in the prescribing of antibiotics.9 All told, peer comparison nudges might be worthwhile, but we are skeptical that they are responsible for the full magnitude of the effect that we estimated.
New Clinical Information
The opioid overdose letters told prescribers about a patient who recently overdosed, a fact that they may not have previously known; thus, the letters communicated a concerning safety outcome to the prescribers. There is evidence that observing adverse events has a powerful effect on prescribing behavior, but it can be difficult for a physician to learn that his or her patient experienced an overdose.26 By conveying this information to prescribers, the opioid letters may have changed their sense of when the benefits of opioids were worth the risks, leading to a reduction in prescriptions.
Reviews, Warnings, and Incentives
The quetiapine letters stated that the prescribers were under review by Medicare and that extremely inappropriate prescribing could trigger administrative actions such as audits. Likewise, the opioid overdose letters, while using supportive messaging, implied that the medical examiner had linked the targeted practitioner’s prescribing with a patient’s death. Both letters may have changed doctors’ perceptions regarding the incentives of prescribing. This possibility is reminiscent of the findings of previous studies suggesting that penalty-focused messages can make people more likely to pay taxes and fees.27,28 Given the mixed results for peer comparison messages alone, it may have been this aspect of the quetiapine letters that made the difference.
The Key to a Successful Intervention
While both the opioid overdose and quetiapine letters could have influenced physicians through many channels, they shared the feature that their messages were surprising: they provided information above and beyond what the physicians already knew. Taken together, these findings suggest that the key to a successful letter is a message that changes prescribers’ expectations.
Limitations of the Interventions
The two recent studies mentioned above15,19 suggest that going beyond a simple nudge can have big effects — so should health care stakeholders plan on barraging prescribers with tough letters like the ones we have described? Although we are enthusiastic about the potential for well-targeted letters to have a positive clinical impact, this approach has limitations.
First, letters with negative or punitive messages can be relatively blunt instruments. In our study on quetiapine, physicians cut back on appropriate prescriptions as well as on inappropriate ones.15 Although we did not observe any worse health outcomes for patients, it is possible there may have been negative effects that we did not measure. Fine-tuning letters to specifically target inappropriate prescribing will be an important step for future work.
Moreover, the messages in these letters were attention-getting. There is no guarantee that similar messages would be as effective in the future, particularly if doctors were to become saturated with messages — even ones that might initially have been shocking. Organizations should think of the utility of such letters as an exhaustible resource to be conserved for critical issues.
We also acknowledge that these findings on attention-grabbing messages are based on just two research studies. It is possible that messages targeting other forms of low-value health care, using different language, or coming from other stakeholders will yield different effects. More work is needed to establish the factors that determine when such messages are effective.
Implications for Practitioners
Our experience in conducting this research and in following other recent developments in stemming overprescribing has led us to recognize several takeaways for practitioners seeking to emulate this work.
Although nudge interventions and strongly worded letters are not a universal solution to overprescribing, they clearly deserve a place in the toolkit for insurers, government agencies, and delivery systems seeking to improve the quality of care for patients — or simply seeking to reduce waste. And while there is still a role for expensive and time-intensive interventions such as audits, investigations, and training programs, we believe that a relatively low-cost, low-touch approach can work surprisingly well.
Practitioners likely will find that these interventions need to be adapted for their local contexts. For example, a postal mail intervention might be inappropriate for a health care organization that utilizes an electronic medical record (EMR) system. Strong messages could instead be delivered over email or within the EMR system, particularly when they contain time-sensitive information.
In addition, practitioners will have access to different, potentially less-comprehensive data compared with the data used in previous letters. Yet the EMR data that practitioners often access can also be much more specific than the administrative data used in previous research. Stakeholders can use these detailed data to closely target interventions and tailor messages, addressing the concern that previous interventions were too blunt and risked unintended clinical consequences.
Lessons for Policymakers
Our research partnership has helped CMS build evidence quickly on what works to reduce overprescribing of powerful drugs. Our research on quetiapine was part of a larger project to address overprescribing of powerful drugs in Medicare.29 By harnessing the wealth of timely data that CMS automatically collects, we were able to rapidly test, evaluate, and revise our letters. Given the ubiquity of such claims data, a similar rapid-cycle testing model could be attractive to other state and federal health agencies. And this approach is not limited to government agencies: health insurers and health systems in the private sector also have access to similar data and could employ a similar strategy of evaluation and testing.
We also noted a key, unique advantage of the opioid overdose letters: they provided information to physicians — overdose death records — that normally would be hard to access. Going forward, some of the most promising avenues for stemming overprescribing involve policies designed to make it easier for clinicians to access useful information. Prescribing records from state prescription drug monitoring programs, health insurance claim records, and inpatient and emergency department encounters collected from hospitals are all too often siloed, stored at different state agencies or with different health insurance companies. Merging these databases will improve public health surveillance, and the combined data could be of tremendous benefit to clinicians.
Much of the new and promising research on overprescribing involves government-academic partnerships. This fact highlights how governments and researchers can work together to create policies that address public health issues and improve health outcomes. There are many opportunities in government to randomize policy interventions as they are rolled out, with minimal extra effort. For agencies interested in increasing the quality of care but short on resources and funding to randomize and evaluate new approaches, partnerships with academic researchers could fill the gap. Organizations such as the Office of Evaluation Sciences, J-PAL North America, The Lab @ DC, and RIPL (previously known as the Rhode Island Innovative Policy Lab) are facilitating these connections already.
The Power of Simple Messages
With careful design, behaviorally informed nudges and related messages could be a key tool to curtail waste and improve quality in health care. Recent research indicates that small interventions can play a powerful role in shifting physician behavior. We hope that the body of evidence encourages policymakers and institutions to consider these promising new approaches to reduce overprescribing more generally.
The authors thank Spencer Crawford, Anna Spier, and Bridget Wack of J-PAL North America for their assistance with this article.